Cheshire Cat Computing

Hi,

I am having a problem with MRTG when monitor numerous devices. If one of them fails, it causes all graphs to fail. I have set SNMP options to retries 2 and timeout 2. Everything is running smoothly until some device goes down for some reason.

After that I get an error message to the log, which indicates that MRTG wasn't able to get data from device. that is OK
2007-01-28 22:55:54 -- Exiting subroutine via last at c:/mrtg/bin\..\lib\mrtg2/Net_SNMP_util.pm line 1780.
2007-01-28 22:55:54 -- Use of uninitialized value in concatenation (.) or string at c:\\mrtg\\bin\\mrtg line 2113

and so on.

How to I make the MRTG to continue from this to other devices and continue poll other devices. It seems so that the MRTG process gets stuck in "a loop" with the non responsive device and does not know how to go on with the ones that are working.

Any ideas how the change the behaviour of the MRTG on these kind of thing. Currently I am manualy removing the device from poll list until it starts to response again.

Other problem is that I get a lot of errors
2007-01-29 09:32:52 -- 2007-01-29 09:32:52: WARNING: data collection did not complete within interval!
I have a lot of devices running with interval 1 min and rest of them interval 2 minutes.

I am running the MRTG with RRDtool and routers2cgi on win2k3 box

Thanks

1 min is a very short interval - remember that all devices and interfaces are polled sequentially, so you may not have time within 1 min to poll everything.

Also, if you have some 1-min and some 2-min, remember that you need a different instance of the MRTG daemon for each set since the daemon assumes all entries have the same polling frequency.

If you are using UNIX for your server, then you should try running multiple MRTG instances via cron (rather than via the -d option) so that a failure will not affect the others. Also, this way allows you to multithread the checks and you can process more at once.

If you still have performance issues, then yuo should probably reduce your polling interval to 5min. Other ways to speed things up include getting the fastest SCSI disks you can, getting more memory (if you multithread the data collection) and so on.

_________________
Steve Shipway
UNIX Systems, ITSS, University of Auckland, NZ
Woe unto them that rise up early in the morning... -- Isaiah 5:11

Thanks for the answer.
Unfortunately I am running MRTG on Windows2k3. Is there a way to run multiple instances in windows.
I am currently running the MRTG as windows service with one single include file. In this include file I have all the devices listed.

So should I do it like mrtg1.cfg for 1 minute interval devices and mrtg2.cfg for minutes devices and so on.

Well how do I run these at the same time on windows ?

The 'single include file' method is the simplest, but it has the disadvantage (as you have seen) that it is single-threading the queries, and will stop processing if it finds a cfg file error (although it will continue if it is just an SNMP error it encounters).

I would suggest dropping toa a 5-min poll cycle unless you really need 1-min else you'll just not have enough time to complete the poll.

If you absolutely must have 1-min then either get a faster machine, or multi-thread. Since I'm a UNIX guy I dont know how you'd go about doing this using Windows services... maybe another Windows person here can help? Also, try reading the help on http://www.mrtg.org and see if there are more hints there for Windows users.

_________________
Steve Shipway
UNIX Systems, ITSS, University of Auckland, NZ
Woe unto them that rise up early in the morning... -- Isaiah 5:11